These requests came in over the transom in the last couple of days. Maybe some StorageMojo readers have wisdom to share.
I have a question I hope you can help me with. My boss asked me . . . to research HP Left-hand SANs and Dell Equallogic SANs. Do you have any special knowledge of these products and, if so, would you make an informal recommendation?
What say you, StorageMojo readers? If you evaluated both, why did you make the choice you did? Vendors welcome to comment, but please identify yourself as such.
The StorageMojo take
AFAIK, both products are good iSCSI systems. Both are backed by major corporations. EqualLogic may be stronger in the channel today, but HP has channel chops as well. HP’s blade servers may be a more expandable platform, but EqualLogic’s software portfolio may be more affordable.
Translation: you could do worse than either of these.
Part II
Another customer perplexity: service.
We have a pair of HP disk arrays, EVA 8000 and 6000 and I am looking for a consultant to help up with storage planning. Do you do such work or could you recommend someone to me. I am looking for someone who goes beyond just being a seller, I have plenty of potential sellers already.
The writer is in a small city in the Mountain West, so you should be used to working remotely with clients. No, not in Arizona.
The StorageMojo take
HP folks may be wondering: why doesn’t he call HP? My guess: not big enough for a direct engagement.
Courteous comments welcome, of course.
Well. The main deal with Equalogic, in comparison to Left Hand, is that EQ does block-level tiering across media types and RAID levels, where Left Hand does not. Also, EqualLogic is fully clustered, redirecting I/O’s the the nodes where the data is, and this will include having full LUN’s spread across the whole cluster and load balanced. Left Hand doesn’t do that. Really, in all my studies of the storage industry in the last 3 years, I have found no vendor with a better iSCSI system than EqualLogic.
For clarity, however, we own quiet a lot of EQ, but no Left Hand. Perhaps someone could chime in who made the comparison and went the other way? I do know that LeftHand was strongly selling the idea of harvesting DAS from ESX hosts into a usable SAN at one time. There could be an angle there; I just never went for it, because I wanted the storage systems distinct and separate from the compute cluster.
Joe Kraska
BAE Systems
San Diego CA US
I’ve got Lefthand vs Equallogic material on my blog and site.
Here are some resources: http://www.gcsaustin.com/blog/post/Clustered-SAN-Solution-Comparison-LeftHand-vs-EqualLogic.aspx
Find a comparison matrix here:
http://www.gcsaustin.com/hp_lefthand_sans.aspx (bottom)
The blog also has a SAN buyer’s guide which may be helpful.
http://www.gcsaustin.com/blog/post/SMB-SAN-Buyers-Guide-Part-1-Market-Overview.aspx
http://www.gcsaustin.com/blog/post/SMB-SAN-Buyers-Guide-Part-2-Advanced-Feature-Discussion.aspx
I’m happy to help contrast these two SANs. Price is probably going to fall HP’s way. I sell both the HP MSA and Lefthand SANs.
@Joe – HP Lefthand does offer a Virtual Software Appliance that can be installed on ESX hosts to avoid the puchase of additional storage. Most still buy the hardware platform. Your first purchase with Lefthand provides two complete units, single controller in each in a true cluster. The price for these two boxes is equivalent to an Equallogic box.
Lefthand absolutely spreads LUNs across nodes. The SAN/IQ Network Raid functionality allows RAID arrays to easily span multiple nodes. It might be said that with a standard build of two nodes it shows Equallogic what clustering is about.
Both are nice SANs, no doubt. At a low to mid level entry point the Lefthand beats up on Equallogic.
Any thoughts adding Nexenta to this comparison?
We chose LeftHand for two reasons. The redundancy model is much better (lose any N nodes completely), versus the equallogic model of proprietary redundant controllers in special boxes. You can’t offline a chassis with equallogic clusters.
Second, we denied both extensively in-house, and performance was mostly the same for the same price points. Except when snapshots were enabled. LeftHand was approximately 3x better random write IOPS than equallogic with snaps on. And equallogic snaps required about 20x the space. Both problems seem to be caused by a very large internal block size used by equallogic.
Oh, and LeftHand is cheaper by 10-15% for the same capacity and iops.
I’ve been an Equallogic customer for the last couple of years, but have come to a point where Equallogics model, where capacity and performance are increased with a new box containing disks and controllers, are becoming expensive in terms of service and sw maintenance. As a product management, availability, performance, monitoring and licensing is just a perfect fit for us. But the product has yet to be improved feature wise. It’s the same products as of 3 years ago (ok they added RAID6 and got replication to be easy), I miss features like dedup and other technologies that will help in a lot of business or optimizing features like the one Compellent have.
I wish that vendor roadmaps where more available. Perhaps one day StorageMojo would also have a roadmap section like the pricelist section.
Commenter Joe Kraska claimed: “Also, EqualLogic is fully clustered, redirecting I/O’s the the nodes where the data is, and this will include having full LUN’s spread across the whole cluster and load balanced. Left Hand doesn’t do that.”
That claim is 100% false. Spreading LUNs across all nodes is exactly what Lefthand does. Add a node, and you can get more space and IOPS, or more replicas of you data. You can also choose two, three, or four-way replication on a per lun basis, depending on how critical that data is.
Equallogic’s GUI is certainly the easiest to use of any SAN vendor, but the LeftHand tools are certainly not hard to use.
Equallogic solutions are far more space and energy dense than Lefthand, which are based on standard servers. LeftHand is about 6 spindles per 1u of rack space. Some equallogic models offer 2x that density.
A few facts worth noting…
EqualLogic was rated #1 in latest Storage Magazine survey of mid-range SAN users (out of a field of nine in which LH wasn’t represented – apparently not enough users responded). This included #1 to the ‘would you buy this solution if you had it to do over again?’ question.
Gartner Market Scope for iSCSI SANs gave ‘Strong Positives’ to just two vendors: EqualLogic and NetApp.
Dell paid $1.4 billion for EqualLogic less than two years ago while HP paid <$500 million for LeftHand a year later (why did Dell pass up LH in the first round – it wasn’t because Dell wasn’t familiar with LH as the latter developed SAN/iQ on Dell servers).
EL customers numbered 3,300 when acquired; today, according to recent public statements by Michael Dell, that number is over 10,000 today. Does HP release LeftHand’s customer count?
LeftHand relies on redundant hardware resources to provide data protection: either 2-way Network RAID on top of RAID 5 in storage module = 35% usable (10TB raw = 3.5TB usable), or 3-way nRAID, which translates to 24% (10TB = 2.4TB usable). Most all EL users go with either RAID 50 or RAID 5. This constitutes a huge difference between the two solutions in terms of USABLE Terabytes of storage per dollar.
ALL EqualLogic platforms – from the PS50 14 drive chassis introduced several years ago to the 48 drive PS6500 chassis introduced this year – are fully interoperable and run on the same firmware image; LeftHand users mix different generations and hardware platforms at their own peril (thus begging the benefits of “industry standard†components).
EqualLogic has consistently maintained a leadership position on VMware’s HCL and along with EMC and NetApp is one of three Development Partners with VMware. VMware has been using EqualLogic in their development labs since way before the Dell acquisition.
EqualLogic storage is not a cluster insofar as there is no shared memory, cache or processors. It makes more sense to think of it as a distributed volume manager with redundant controllers in each node. Their nodes have redundant components, which makes them less likely to fail than an HP server. Details of failure mode performance should be considered carefully.
@Billy The usable disk space numbers you reference are a joke! Of course there is less usable disk space if you’re in a truly clustered environment. You can choose to use Network RAID per volume on the Lefthand. The only time you’ll give up that space is by choosing to do so. What if you choose to do so on a entry-level EL buy? You can’t. Not even an option.
re “EQ does block-level tiering across media types and RAID levels”
Huh? EQ provides one RAID-level per box. You can migrate from RAID-50 to RAID-10, once. Not the other way, of course! I imagine you can migrate a volume from a RAID-50 box (or set of boxes, as they stripe on top of these) to a RAID-10 box(es), but this migration doesn’t get more granular than an entire volume.
@Joe – If you want to gain the benefits of a cluster in LH you need at the minimum RAID at the hardware level and RAID at the cluster level, so having two running RAID protection schemes will of course decrease RAW vs. Available capacity, this story gets better with Network RAID as you increase the # of nodes in an LH cluster.
@Art – Equallogic can in fact spread a volume across disparate RAID types at the “block” level, but will not move data around once written unless done via a volume re-distrabution, but will allow writes to differently configured “memebers”. I say “block” in quotes because it isn’t a true block, but something closer to a “chunk or page”.
In general, both are great platforms and have really made the iSCSI market what it is today, but both need to take the next step to supporting faster transport technologies, however, LH will always be limited by the “server” platform, so 10GigE might not be of any benefit, even though it can be easily brought to market in a commodity based platform. Equallogic, which I believe is limited by the 1GigE connectivity, has been slow to 10GigE support since they will need to spin new controllers to support the new transport technology.
The VSA is a cute application, but is clearly NOT meant for production storage requirements, how many virtualization layers do you want your IO to pass through?
At the end of the day, both have merits, both aren’t going away, and both will work. I’ve written before about how “server” based controllers and “custom” controllers are pretty much equal when a company is creating a new product, both have merits and both have deficiencies.
@Joe – The LeftHand architecture is based on servers that have single points of failure and insufficient cache protection. Therefore one sees in the LH SAN/iQ user document that they have two best practice recommendations depending on whether a user’s data is “mission critical†or “criticalâ€. From page 268:
Best Practice for Setting Replication Levels and Redundancy Modes
For mission-critical data and using a 3-node cluster, choose 3-Way or 4-Way replication and redundancy priority. This configuration sustains the first fault and ensures that the volume is redundant and available.
If your volumes contain critical data, configure them for 2-Way replication and a priority of redundancy.
(Note: EqualLogic treats ALL data as mission critical).
re “Equallogic can in fact spread a volume across disparate RAID types at the “block†level”
Are you saying within a volume my data can be haphazardly distributed between RAID-1 RAID-5? Sounds terrifying and full of performance problems. I’m not sure I believe EqualLogic would let you do this. Can anyone else confirm the Technologist’s claim?
“Are you saying within a volume my data can be haphazardly distributed between RAID-1 RAID-5?”
I don’t think “haphazardly” is a good description, but yeah, Equallogic will by default distribute volume content across different RAID types. My understanding is that it actually makes decisions about what blocks to place where based on activity. More frequently access blocks end up on faster members. If you want, you can fix a volume to a specific RAID flavor.
Ok, let’s explore the facts:
EqualLogic can move volumes from one RAID type to another, but if you have multiple RAID types in a pool you’re limiting the number of spindles per RAID type. For example, a pool of 32 disks (2 nodes) in RAID 50 is faster than 16 disks in RAID 10 and 16 disks in RAID 50, so why not have everything set to one RAID type? In other words, if spanning the maximum disks available is always faster, then EqualLogic’s data tiering between RAID types is useless because it always gives you less than the maximum spindles that are available in the pool.
If a single EqualLogic box fails, all volumes spanning that box are down. With LeftHand’s Network RAID, the SAN can sustain multiple disk faults across multiple nodes, complete node faults, and complete site faults without volumes going offline. EqualLogic has no way of doing synchronous replication between racks in a data center, floors in a building, or across a campus network. This means EqualLogic doesn’t have an HA solution that delivers RPO=0, RTO=0. This SAN functionality is a requirement for VMware HA and Fault Tolerance features.
LeftHand’s usable capacity hit from Network RAID is what it is, but LeftHand eliminates reserves for thin provisioned volumes, snapshots and remote replicas, which nets you the same or better utilization as EqualLogic. EqualLogic allocates a 10% buffer for thin provisioned volumes, minimum of 10% for snapshot reserve, 10% for Remote Replica reserve, and 10% for remote replica failback snapshot reserve. Bottom line: EqualLogic’s usable capacity at the primary site is about 39%.
EqualLogic also uses a 15MB page size compared to LeftHand’s 256k page size. This means a single block change in a volume can result in a 15MB snapshot. This is why EqualLogic’s best practices are to maintain a Snapshot reserve equal to 100% of the volume size. When doing remote replication, which is based on snapshot deltas, you end up moving up to 58 times as much data over the wire.
Yes both solutions can span volumes across multiple units, but EqualLogic is limited to 8 nodes for volume spanning and 12 nodes in the overall solution. LeftHand can span volumes across up to 32 or more nodes (no hard limit), and no hard limit of total nodes in a single management group. LeftHand also has a patented plug-in for MPIO that delivers data locality awareness at the client and allows you to maintain multiple, simultaneous iSCSI sessions to the nodes. This eliminates the I/O forwarding hop on the backend that EqualLogic has.
Last but not least LeftHand runs on open x86 enterprise class storage server platforms, from blades to rack servers. This makes the adoption of new hardware technology like 10 Gig, Nehalem processors, PCI Express RAID controllers and faster memory much easier. EqualLogic is tied to a proprietary Broadcom MIPS processor chipset that significantly underperforms the latest x86 server hardware, which is why EqualLogic is taking forever to deliver 10 Gig. Could EQL move to x86? This would require a complete firmware re-write, because their cache mirroring firmware is tied to the Broadcom hardware.
LeftHand’s VSA runs in a Virtual Machine today, but could be native tomorrow. This is a great solution for remote office delta replication back to a central site. With EqualLogic you have to put hardware at all the remote sites to deliver the same functionality.
Aside from all this both products are very good and will satisfy most customers’ SAN requirements. The real kicker for EqualLogic and LeftHand is the fact that both products outperform and are much easier to manage than most of the legacy scale-up Fibre Channel solutions out there. Both solutions also come with all software features included.
[Ed. note: John is LeftHand’s former CTO, now at HP.]
Very interesting to read, all good posts and comments here about LeftHand and EqualLogic.
I especially liked reading what John Spiers said as I have always been a fan of LeftHand due to its support of open industry-standard hardware and ability to have unlimited storage clustering that scales out beyond just two storage nodes. I think that this is the right strategy and we espouse and support this same “industry standards” philosophy here at StarWind.
It’s a significant benefit for customers to have storage like LeftHand which runs on any standard, enterprise-class x86 and x64 storage server. Thus upgrading to new hardware is dramatically simpler than using proprietary solution. Allowing customers to buy widely available standard servers with full redundancy and hot-swappable components means that they choose from Dell, HP or IBM or others without being locked in although after HP acquisition LeftHand only sells on HP ProLiant which is a Cadillac of servers.
We also found that this freedom to choose from any standard hardware is important to small and midsize companies who want to deploy networked storage cost-effectively, quickly and easily without the interoperability issues and high prices that are associated with proprietary storage solutions.
(In the spirit of full disclosure: I am CEO of StarWind and we develop open iSCSI SAN software which works similarly to SAN/iQ software from LeftHand.)
@John S. (note, I was previously with Equallogic and Dell, and John and I have talked about product differences before).
“If a single EqualLogic box fails, all volumes spanning that box are down.”
-This would be the same as a LH cluster going down. Each Equallogic “member” (box) is built for 99.999% availability, at both the hardware and software layer. LH has never had ANY control over the HW layer, so any proposed “uptime” is 100% based on the LH OS and software layer. I will concede that LH does however have the ability to do “campus clustering” but at the added cost of 2x the entire solution for redundant HW and Software.
insert comment about usable capacity. I will absolutely take total usable capacity of Equallogic over LH any day of the week. Reserves are still areas available for data, LH’s layering of software RAID on top of a servers HW RAID will always use MORE space for simply RAID protection and Cluster protection. In a 2 node cluster LH requires 100% mirrored data for an HA configuration.
“When doing remote replication, which is based on snapshot deltas, you end up moving up to 58 times as much data over the wire.”
-Remote replication is based on a 64K block size. Remote replicas are also thin provisioned by default. (note: replication has always been based on a 64K block size) I will concede that Snapshots are based on the 15Mb page size, however, considering the functionality of EQL’s snapshots over LH (LH has a grandparent, parent, child relationship which will force the maintaining of old snapshots in the middle of a snapshot tree, as well as utilizes a “copy-on-write” technology that has huge performance implications.)
“LeftHand also has a patented plug-in for MPIO that delivers data locality awareness at the client and allows you to maintain multiple, simultaneous iSCSI sessions to the nodes. This eliminates the I/O forwarding hop on the backend that EqualLogic has.”
This feature exists for the DSM on windows for Equallogic as well. Storage groups are now increased to 16 members as well. Agreed about LH’s ability to span more “nodes” however, since LH’s cluster is an N+1, the more nodes you have the greater the risk of multiple node failure.
“EqualLogic is tied to a proprietary Broadcom MIPS processor chipset that significantly underperforms the latest x86 server hardware, which is why EqualLogic is taking forever to deliver 10 Gig. Could EQL move to x86? This would require a complete firmware re-write, because their cache mirroring firmware is tied to the Broadcom hardware.”
-Can’t really disclose too much here, but the 4 1GigE ports on a single member from equallogic has been proven to beat the performance of any number of 10GigE ports on a 2-3 node LH cluster, Intels processor and Dell/HP servers were NEVER built for IO. Equallogic has a purpose built HW platform and a purpose built multi kernel OS built for iSCSI networking and IO processing.
VSA is pretty cool, basically takes the LH OS and makes it a virtual machine. Glad it wasn’t really very prevalent when I was selling into SMB a couple years ago.
Reality, both LH and EQL have been VERY successful in the storage market, mostly because of stuff NOT discussed in the comments. They both hands-down out-perform the iSCSI implementations of competitors. They both are MUCH easier to manage and implement then almost any storage platform on the market. iSCSI is night-and-day easier then FC to work with, and doesn’t take a “special” skill set or professional services engagement to get running like a well oiled machine.
Regarding how much IO bandwidth you can get out of an x64/Windows box. . .in my experience, 4Gbps from disk to network (via teamed gigE ports) is doable on Intel 5000P chipset / 8-core servers. Up to 6Gbps from RAM. 15+Gbps is doable on a Nehalem/Win2k8 system, though we have found stability problems with current drivers (Intel NIC, Supermicro mobo).
@John Spiers’ (of HP, former LeftHand CTO) post
Up-front disclosure: I work with Dell’s storage group.
It’s flattering that Mr. Spiers spends so much time thinking about EqualLogic. His reply actually contains many inaccuracies about EqualLogic products, including comments on IO forwarding, multipathing, software portability and chip set support, space allocation, replication bandwidth, VMware HA and FT support, and EqualLogic scaling limits. If you are in IT and are weighing storage options and find any of these topics to be relevant, please contact Dell for the correct information about EqualLogic. We’d love to talk to you about the benefits and true capabilities of our products.
Incidentally, it seems inconsistent for Mr. Spiers, who emphasizes reliability as a key benefit of LeftHand, to also suggest scaling LeftHand clusters above the limits recommended by their own user manual.
He also brought up a number of technical points which can be better understood at a conceptual level by comparing what I’ll call Purpose-Built-Enterprise-Storage (PBES) and LeftHand Clusters (LHC):
1) All physical disks still accessible if a controller is offline :
PBES: Yes (advantage); LHC: No
2) Application recovery manager, not just a VSS snapshot requestor:
PBES: Yes; LHC: No
3) Copies recommended for critical, high-availability data:
PBES: 1 (2 to synch two sites); LHC: 3 (4 to synch two sites)
4) Stripes data across multiple pairs of components; one of each pair must be
online for the volume to be online:
PBES: Yes; LHC: No (which Mr. Spiers emphasizes).
5) No-single-point-of-failure built into one box:
PBES: Yes; LHC: No
6) HP’ highest offering for mission-critical data:
PBES: Yes; LHC: No
To be clear, the above discussion relates solely to HP products, with the purpose-built enterprise storage being the EVA and XP (XP-only for #4,5,6).
As for EqualLogic, it’s purpose-built enterprise storage, but with a virtualized and intelligent scale-out architecture. If you’d like to use synchronous replication software with it, Dell sells and supports it.
In considering your SAN solution, we suggest that special emphasis be placed on application recovery since many more outages are caused by a virus or a human error than a failure of an enterprise class storage array.
In summary, if you follow best practices and consider business-relevant risks, you’ll see that EqualLogic has advantages for SMBs and enterprises alike, as well as IMHO being a simpler, cleaner, more power-efficient, and denser solution.
— Rob Young (at Dell) —
Great respones from everyone. Not much to add except the fact that a lot of th epostive stuff with Left Hand around open standars will likely be limited soon as HP is adding their traditional propriatary staregy of locking customers in. The exact wat they are doing with VIrtualConnect blade technology. I have used both EQL and LeftHand and to me EQL is a true enterprise class iSCSI alternative to EMC, Hitatchi but with fantastic usability and ease of use. Left Hand on the other side is great storage for SMB, test environements and customers who don´t have cirtical data or applications. And in January EQL is laucnhing a 10Gb iSCSI solution which will eliminate the need for FC for anything.
Great exchange.
Disclosure: As an EqualLogic partner for 4 years (way before the Dell acquisition) we’ve competed many times with the Lefthand product. We’ve done over 50 EqualLogic projects and do not sell HP.
In comparisons based on performance, we won with EqualLogic every time.
The one time Lefthand beat out EqualLogic was for a specific project because the client required synchronous data replication and the Lefthand solution (4-way) provides it. Unfortunately, the customer was sold a 2-way between data centers with a Virtual Software Appliance to handle the 3rd leg of the cluster.
When Dell embeds synchronous replication into the EqualLogic it will take away the single biggest selling point of the Lefthand system.
EqualLogic is also less expensive if compared on an apples-to-apples basis.
Get Self-Service Pricing on the EqualLogic gear here:
http://www.4equallogic.com
As a Lefthand/HP partner today, and a past of working in a rather large Equallogic implementation, I’d be very interested to know when Equallogic would include synchronous replication that’s actually possible to set up.
Our customer (that still runs Equallogic) has around 150 VMs running on XenServer, with the Dell Equallogic SR adapter, meaning each VM has at least one LUN, most have two, on the Equallogic group.
Currently, setting up even asynchronous replication consists of having to create a replication schedule MANUALLY for EACH LUN, and setting up replication manually on EACH LUN, one by one.
Clearly, this is nearly impossibly inefficient for this kind of setup, and I find Equallogic currently lacking badly in the administration department when you get to this level/number of LUNs and clients.
@Henning
You can create a Collection in the EQL PS Manager GUI which will then allow you to create your replication schedule on those volumes within the collection.
This is an issue more so thanks to Citrix – each VM requires its own volume on the SAN, and thus if you have 100 VMs, you need to configure and replicate 100 volumes. Hyper-V and VMware both use shared volumes which allow for easier replication but also quicker points of recovery as you have a smaller number of volumes to bring back up.
The main differential between LH and EQL is that LeftHand require a cluster of exactly the same nodes (disk sizes, types and hardware generations). Thus if the next generation of product then gets released, you have to build a new cluster, meaning you’ve now got two seperate islands of storage to manage. In an EQL world you can have every single product ever made in the same virtual SAN with volumes spanning and load balancing across them – they all run the same standard 20mb firmware with no restrictions.
At the end of the day, they’re both great products which both have their strengths.
We’ve had first hand experience with both LHN and EQL.
I really like the LHN product on paper. We’ve had it deployed for a couple of years (fairly big deployments in the US, Europe, and Asia). LHN is pretty rock solid overall, and they were reasonably pleasant to deal with. I really liked a few things about them:
1. It runs on commodity hardware. We have several dozen nodes running on Dell iron.
2. They have a sensible ‘any given node can fail’ approach to clustering. We’ve taken advantage of this several times and upgraded an entire cluster node by node without taking storage down at all.
#1 was a big advantage for us, especially because we are a big Dell shop. LHN being able to run (and be supported) on our standard x64 commodity Dell servers (PE2950) was a _big_ plus. In addition, LHN promised us Xen support for the VSA (as opposed to VMWare). Also a big plus (especially before ESXi).
That was then. Pre-HP.
This is now. Post-HP.
First of all, the quality of support at LHN took a sharp dip post-HP. We are barely able to get an acknowledgment on problems that make the API unusable for our needs.
Most importantly though, they are no longer an ‘Open Platform’. They have pulled support of Dell hardware (my biggest fear of the HP acquisition). Going forward, you will only be able to run LHN on HP hardware.
Lastly, they don’t appear to have any plans to support Xen any time soon.
No hard feelings against LHN. They had/have a good product. But due to the above we have started (slowly) replacing all our LHN with EQL.
So far so good on EQL. Although personally, I wish Dell had bought LHN instead.
@RD…
I have had exactly the same experience. As a customer of both, we too are moving to EQ fully eventually. I told my Dell account manager many years ago that Dell should buy LH.
We had a major SAN outage with LH about 8 months ago. We were using all of LHs best practices and still we had volumes go offline for 4+ hours after the failure of one node. The biggest kicker is that when working with IBM (our nodes run on IBM hardware) to solve the hardware problem, IBM was less than estatic to work with HP and I had to manage both LH and IBM to solve my problem.
Our LH SAN is one of many makes and models..Yes it is cool that it runs on x86 but from week one to week two you don’t know what you are getting from LH. First it was the SuperMicro servers, then IBM, then Dell, then HP. Oh yes I have all of them…each different hardware RAID cards, drives, power supplies, etc
Our newest HP stuff looses a drive a month. I wish I could run LH’s management console on Dell’s hardware with Dells Call Center for Support.
I do think the EQ is like running SAN IQ 6.5 as far as features and functionality.
A bit late to the party, but wish to share my experiences.
We use LH and EQL in our shop, and both are great products with distinct strengths and weaknesses.
Lefthand is a much more scalable platform, as you can add storage in “bricks” to expand capacity and performance. They’re very easy to set up, and installation is predictable and consistent. However, the commodity grade hardware they’re built on is subject to all the commodity grade failures. If you want to have a chance at implementing a highly available SAN you MUST have at least 2 nodes and 2-way replication. Every node WILL have some sort of failure that takes it offline. We have also experienced an issue with LUNs going offline without warning, and on HP-branded hardware. Their software is really good though, and performs reasonably well. I think the HP acquisition has hurt the product line more than it has helped so far. The integration of the support site into the HP site has been a nightmare, and whenever LH techs need to call HP hardware techs to get a part replaced it takes forever, and is far from a smooth transaction. At this point the current LH offering is a bit long in the tooth. Not much has really changed in over a year now. No new features, no new functionality, no new major releases. In the end, LH is suffering from the same things that plague nearly ALL HP acquisitions. HP has become what IBM was 20 years ago. A big, slow moving monstrosity with overpriced products and poor support.
EqualLogic – I don’t know what others are paying, but I get more useable TB for the $$$ than with LH. My PS5000 and PS6000 arrays have worked almost flawlessly, with the exception of a firmware issue we had a few months ago, which was quickly dealt with. None of the arrays has ever had a “down” condition (we have 11 of them). The redundancy model they use is different from LH, and it works. They employ the same kind of model as your typical Cisco Catalyst switch – redundant components, shared chassis, and more controller cache per array than LH. Since the start of this blog, they have also started to offer 10Gb options, and surprisingly they don’t add a lot of cost to the array – the 10Gb switches is where you’ll pay thru the nose. We have just purchased some PS6500 arrays (the 48 drive units), but have not yet put them into production. So far, all I can say about them is that they are a royal pain to rack, and the instructions are horrible. But I expect that once they’re in they will meet my expectations.
I have heard that LH’s SAN/iQ 8.5 and 9.0 will be a huge improvement over their current offering, but that remains to be seen. Today, EQL has the upper hand in the management and feature department, and show no signs of slowing down anytime soon.
I researched both heavily, and went with LH mostly for the redundancy factor. If you don’t replicate a volume across units, you end up with the same basic useable capacity as an EQL unit. Sales people want to push the storage loss for using network RAID; to me, it was a feature that I wanted to implement and EQL couldn’t provide.
99.999 was mentioned over and over by my rep, but five nines on a piece of paper doesn’t make me feel good when the backplane blows on the hardware. Is it probable? No, but I sleep better at night. I have tested node failure on the LH product by pulling out all the network cables on a node and the volumes stayed online. You can’t do that on EQL at the entry level.
Plus LH is cheaper.
I have also since grown to appreciate the LH VSA. It’s actually free to use as a basic iSCSI target software and can come in handy for cheap, low-performance storage needs. For instance, I used it as a deduplication store for VMware Data Recovery and it does a great job.
I spend lots of time about four years ago looking at the various iSCSI vendors and quickly narrowed it down to NetApp (because we had lots in-house), StoneFly, LeftHand and Equallogic. NetApp was the most expensive, had issues with the “forklift head upgrade syndrom” and wanted to nickle and dime you on features (forget to ask for replication? That’s another five useable terabytes in licensing fee’s vs. other vendors). StoneFly looked interesting but didn’t seem robust – their bankruptcy (and subsequent re-emergance) was a little unsettling. LeftHand was my strong favorite for a while. My only concern was their commodity grade hardware. Intel servers aren’t known for their massive I/O. Then almost at the last minute, someone recommended Equallogic. They were quick to bring a unit out and do a full demo. The management interface blew me away, and I liked the attention to detail on the special-built hardware. They followed through with their promise of future compatibility and I think that’s somnething that should not be trivialized or overlooked – my units that are now over five years old (time flies now that I think about it) will integrate just as seamlessly with the latest models. That’s the advantage, to me, of their special built hardware and controllers.
“commodity” hardware is only an advantage if it brings tangibile benefits and advantages over other solutions. If it is true that post-HP aquisition that LH will only run on HP hardware then that just killed their biggest advantage in my mind. I have other units within my organization and in sister organizations that are running Equallogic and I have yet to hear anything but praise for them. With the renewed focus on energy effeciency and power useage, their advantages there are also very compelling.
I am glad to see they got the 10 gig ethernet out. I don’t work directly with our SANs as much as I once did, but I will be looking for performance data on them and looking forward to being able to finally make the case for halting future procurement of fibre channel.
Eric E
StoneFly(iSCSI.com) , LeftHand and Equallogic ..they were all purchased by bigger companies. because they were all having hot technologies and some sold in $billions or hundreds of millions. With your logic all went bankrupt.
The rest of your comments follow the same logic ..no real research and understanding or any of the products and their ++ or — .
Worked with EQL and Lefthand and currently only working with LH as we found out the the whole replication/reservering snapshoting in LH is better than in EQL. I guess EQL has a better IOPS ratio and maybe a better peformance ratio but choosed for the HA of LH.
@ roland
You state that LH is only for SMB, Test and non-mission critical infrastructures. Did you ever work with LH ? Your opionion is not that objective if you didn’t work with it. If you have a 2 node setup with network raid 10 over 2 sites. Isn’t that mission critical HA? If the backplane of the EQL blows? What happens than?
Read your post with interest. We’ve been running LH since mid 2008, just before the HP takeover, and have experienced a lot of issues. I’d like to tell you our story:
Originally EQL was my preferred choice since before the Dell takeover. Back in 2008 when we were looking at buying our first iSCSI SAN We were hoping Dell’s EQL aquisition might drive down their prices a bit, but this turned out not to be the case, They were just a bit over budget for us at the time.
LH came in a fair bit cheaper for the solution they were offering us back then. The RAW 4.5 TB in a pair of NSM2060’s with 6 x 450GB 15k SAS per node was similar to that availble in the EQL solution. I should add that we were naive in looking at RAW capacity and not sufficiently taking into account the overhead of network RAID 1.
LH also sold us on the concept of their scalable ‘pay as you grow’ approach (grow your storage, Processor, IOPS and Network IO at the same time as you add further nodes in future expansions). We also liked the ability to do network RAID 1 on a per volume basis so your important volumes could stay up even if a node fails.
After almost 3 years running LH, its become clear to us that the TCO for the LH solution is really very high, and the majority of points in the sales pitch have proved very misleading.
Didnt take us long to figure out that RAW capacity figures of a LH cluster are pretty meaningless. We overestimated the reliability of the individulal nodes. We were thinking we could run some volumes without network RAID to maximise space. But really you cant. You really have to run network RAID 1 for everything, because these boxes can – and do – go down, for reasons ranging from hardware failures, to SAN-IQ software bugs. Thus the maximum usable storage with LH should always be in seen in reality as 50% of the RAW capacity at best. The £/TB cost suddenly starts to look a lot less attractive. Your storage costs just doubled.
then the HP takeover happened. We had some big fears about that. Assurances were given to us from LH. they turned out to be pretty worthless. I’ve never been a particularly big fan of HP, and wouldnt willingly be an HP customer. The old LH account management team gave us assurances that LH would remain pretty independent and we shouldnt worry. but:
– the Dell hardware options got canned (we are a dell shop, so this was a big deal)
– support portal went away. we lost access to all the tech info, patches, downloads. HP’s wrote to us telling us that we should start using the ITRC but the lefthand information just wasnt available. It was a complete nightmare. All self help resources just gone, overnight.
– result of the above was we had to uplift our support from 8x5xNBD (which was actually sufficient at the time) to 24x7x365, because we had to call someone if we wanted the smallest bit of info that we used to be able to get from the web.
– when we tried to do this, we found prices for support renewals and uplifts had gone through the roof (50% higher than they were before the HP takeover)
– Where once we had a a UK toll free number to call for support and could always get through to a friendly knowledgable engineer in the US who would resolve whatever issue or queries we had on that first call. This number was suddenly withdrawn and and 1st line UK tech support started going through India. They barely knew what the products were. They also worked on a callback basis. the callbacks would usually come back 3h 50 minutes, so as to just stay inside the SLA, but questions would never be answered on these initial callbacks. The calls would invariably get escalated to the same team in the US, the difference being that issues that took an hour or so to resolve now took several days just to get to the same person we could at one time call direct. A real insult considering we were now expected to pay 50% more for the support contract.
– No account management. nobody in HP would take our calls, listen to out complaints. We couldnt get any product or roadmap information. To all intents and purposes LH had ceased to exist, and we were left on our own.
Fast forward to now, the ITRC is a bit better and we at least have an account manager again, but actual tech support has got even worse. recently we’ve had some serious issues post SAN-IQ 8.5 where solution pack bits started acting up in our environment. Responses from the US team who weere once so good is now totally woeful. Things like ‘try rebooting, uninstalling reinstalling’ We had snapshopts buiding up and up for days because of VSS writer failures. They didnt want to get involved in any root cause analysis. We sent them logs, they wouldnt look at them. They were more concerned in blaming some pretty random stuff in our environment. One tech even asked us to swap out all the patch cables in use in our SAN. We had to turn off VSS writers in the end. HP couldnt find the answer and their closing statement was “we’ve done all we can, we cant help you anymore, we dont know. Sorry”.
Despite all this, we recently we wanted to expand our SAN. We wanted to add more nodes to our cluster and begin to realise the benefits of increased IO, throughput and disk that LH promised. We also wanted to take advantage of the new network RAID 5 level and to try to claw back some usable space in the cluster. (Network RAID 5 currently requires a minimum of 4 nodes in a cluster to make it work)
HP originally lead us to believe that we could add nodes to an existing cluster in pairs – even if they were not the same size as our original ones. They told us that provided we added equal sized nodes in pairs then we could scale out the cluster and get all the benefits. (We questioned that on a couple of occasions and actually have it in writing from them confirming it).
We were about to buy a a pair of P4500 7.2TB nodes when we discovered accidentally from a LH tech, that the information we’d been given by pre-sales was false. We’d have to put these new nodes in a NEW cluster if we wanted to use all the storage. This really means starting a new SAN. If put these new nodes in our current cluster, we’d loose about 60% of their capacity. They actually tried to tell us this wasnt a big deal!
When we bought the 2.2TB NSM 2060’s in 2008 they were at the sweet spot of cost / vs capacity for us. 3 years later, The P4500 7.2TB nodes represent the same optimal choice for us in terms of £/TB. But to grow our SAN the way LH sold it, we would have to continue to buy much smaller boxes at less optimal £/TB, or spend more money to replace the whole lot with the newer nodes. Neither of these choices were optimal, and made a mockery of LH’s ‘pay as you grow’ marketing
Additionally, HP couldnt offer us any hardware that would allow us to expand our current cluster without some wasted space as no ‘exact match’ for the Dell based 2.2TB NSM2060’s exists with HP hardware. So we’d still end up paying for ‘wasted’ space.
as a customer, we had a choice to make. exapnd the current cluster with lower capacity more expensive nodes, with wasted space, eating up rack space and power in the process..
…or, start a new 2 node cluster (really a seperate SAN) with the new 7.2TB capacity nodes.
It seems you can expand your space, OR you can expand your throughout and IO, but you cant do both as the same time at a reasonable price point. Expansion of a LH SAN is always going to mean compromise in this respect, its never going to fulfil its pre-sales promise.
best we’d get for our money was to buy the 7.2TB nodes, create a new cluster, accept the management overhead of managing 2 pools of storage, and sacrifice the scale-up of CPU, IOPS and Network IO and still have no access to the newer network RAID5 features, as both clusters would have 2 nodes each. We also realised that this story would be repeated at the next stage of expansion.
We’ve really regretted choosing LH over EQL back in 2008. It was a bit more expensive initially, but with all the weakneses of the LH solution that have become evident over time, coupled with issues caused by the HP aquisition we’d have been much better off in the long run.
As it turns out, thats what we’re doing. We’ve re-evaluated the EQL kit, and we’ve decided not to buy LH for this expansion. Instead we’re migrating the whole SAN to EQL.
Hope this helps anyone out there about to make a similar choice
@boxing_surfer
We’ve similar problems over the years with companies HP had purchased that almost match your story identically. The latest being IBRIX. But in the past it has taken them about 2 to 3 years to come back to previous support levels and expertise. Here’s hoping LH and IBRIX both get there eventually.
Wow this is an interesting post.
I am evaluating a SAN for to bring our ESX environment into HA/VMotione (and perhaps FT, if we can afford it) territory.
The decision pretty much boils down to LH vs EQ (Starwind VSA on R710s is a distant consideration).
Reading this thread from beginning to end caused me to flip back and forth a few times.
We’re looking at an EQ PS4100XV with async replication to PS4100E versus a P4300 starter SAN kit (2 x DL185 in active/active mode).
After reading all the above, I really like the EQ solution but I’m leaning towards LH. There is about a 20% price difference and LH’s active/active cluster w/ virtual IP is really what we wanted from the get go (high availability without fuss). If my main EQ node goes down, there’s a lot of work to repoint ESX hosts to the secondary (lower performing) node whereas there’s nothing to do in LH.
Any additional comment and/or insight will be appreciated.
Also, what are your thoughts on dedicated switches vs sharing an existing L3, GigE switch using VLAN?
Also, in the event of a primary EQ node failure, I’d lose at least 5 minutes between my live env. and the restored env. for all the data on the SAN (5 mins is the maximum replication interval for EQ). That could be major depending on the application running on the SAN.
@zr07
I believe EQ supports proper multipathing for VMWare environments since 4.x (one of the updates).
If you want to fail over to the async replicated array, not sure that is supported or possible (think cache coherency).
w.r.t dedicated switches vs shared switches, only if your shared switches can do real QoS (need real hardware queues and lots of buffering) on a per VLAN/port basis.
Oops, I didn’t read through boxing_surfer’s post. Wow!
I thought LH was a pretty fair competitor to EQ but those issues are crazy! People typically expect more support for their server, much less a SAN storing tons of their consolidated data. All of the sudden EQ’s recovery plan for the rare total-node failure doesn’t sound so bad.
So I’ve read a few horror stories with LH here (and on other blogs/forums around the web). Any similar stories on EQ? I haven’t seen much thus far.
Dedicated switching is the way to go for sure, and if you’re using the PS6010 series (the ones with 10Gb SFP+ ports), then look no further than the Arista 7124. There are others that will do the trick, but the Arista kicks the crap out of everything I’ve ever seen, and we tested LOTS of switches.
Go with Equallogic; I have been using it for the past three years, 500TB’s, single group, 2 pools. Super fast, easy to manage, great performance management tools and on a network I can manage (ethernet) and optimize easily.
ZRo7,
Note that the EQ systems are almost always dual controller where each LH node is a cascade of single points of failure. Comparing what happens when an EQ fails to what happens when an LH node fails is therefore invalid as the LH is much more likely to fail than BOTH controllers on the EQ are. A single controller failure in the EQ is a non event.
Nothing happens then LH node fails, if you don’t create Network RAID-0.
@ZR07
what did you end up going with?
I just pulled out my lefthand P4000’s for Equallogic PS6100S (2) + PS6100XV-600 (4). Much easier to manage, much, much, faster and more usable storage. I am a former IBRIX customer; I am drinking the EQL kool-aide. This is good stuff.