Isilon increases their IQ

by Robin Harris on Monday, 28 January, 2008

Despite being written off for dead . . .
Isilon’s been putting their IPO money to good use: engineering the next gen of their platform that they’ve named the X-series. In the meantime they’ve been adding customers – over 600 so far – and they have 60 customers running the new kit.

Moving from an aging single-core Xeon to a dual-core Xeon – the second core isn’t turned on yet – with faster busses and more cache speeds things up. They claim up to 60% faster performance, 20% less power and heat and 10 GigE readiness. Once they get their software dual-core aware they’ll have another nice boost to offer.

The StorageMojo take
Turning over the platform more rapidly than traditional array vendors do is a good strategy. It keeps the competition off-balance and gives you something new to tell customers. What good is commodity hardware if you don’t follow Moore’s law?

That said, Isilon’s scale out architecture is the real differentiator vs NetApp and other traditional filers. More bang for the buck just underscores the differences.

Comments welcome.

{ 12 comments… read them below or add one }

tim wissman January 29, 2008 at 5:28 am

the Isilon stuff looks very much like the SpinFX stuff that Netapp procured. from the file system perspective, using something like IBM’s GPFS could be a better suit as well. I have not played with any Isilon stuff, but it looks VERY interesting.

joe m. January 29, 2008 at 10:17 am

I’m VERY interested in Isilon.. If anyone has experience with them please post. Production experience would be best, but I’d be interested in just hearing what you’ve learned from sales visits.

I just contacted Isilon this morning myself, so I will try to share what I learn as this progresses.

TylerB January 29, 2008 at 10:24 am

Robin,
Per our previous discussions on this, I felt Isilon was being too narrowly focused on ‘unstructured data’ (mainly media, GIS, etc.); I think this announcement is big, but not as big as their other annoucement- VMWare 3I certification for the new X series. Ever since I started working with VMWare over NFS, I thought Isilions brick-building style would be suited perfect for for it. I was so disappointed with the engineering response- “oh no, we’re not supporting that kind of transactional data at this time.”

VMware is designed similar- add esx servers and grow your resource pool- now do the same with your storage- add isilon nodes and increase your storage size and performance.

TimC January 29, 2008 at 3:03 pm

@TylerB:

The issue though, is that vmware is EXTREMELY random in it’s data read/writes by nature. Not really the type of transactions/workload that isilon excels at.

The issue I see even now is that isilon is still strictly SATA drives. They typically don’t hold up under high iop workloads, which is exactly what most vmware installs are.

joe m. January 30, 2008 at 5:08 pm

Tyler .. Are you saying the new X-series from Isilon will perform better in a highly random/transactional environment?

I’m curious to hear how Isilon (the new X-series) performs in this type of workload, as this is exactly the type of workload I’ll be talking to them about on Friday afternoon.

TylerB January 30, 2008 at 7:50 pm

joe-

Isilon is claiming improved performance. I can neither confirm nor deny this because I haven’t seen the new boxes nor have I tested them.

TimC-
vmware is extremely random, but it’s low throughput. And most of the vmware installs I’ve seen are not very high iop workloads. The performance bottleneck hierarchy in vmware usually goes- memory, cpu, disk, network. Also as far as SATA, I see a ton of customers deploying vmware on SATA (not isilon per se), but the slower individual speed of each sata drive can be overcome with proper data layout and the appropriate number of spindles. Of course disk for disk it’s slower, but that doesn’t mean you can’t build a solution to support

vo January 30, 2008 at 11:41 pm

We just purchased 6 12000Xnodes after evaluating the previous nodes last year.

iometer show performance of the Xnodes under vmware are 2-5x better than the previous nodes and are inline with the iometer numbers for a Netapp 3070. So 6 nodes will perform about the same a 6 3070’s.

Isilon can dynamically load balance NFS clients on the fly across all nodes. You can setup a cron job to move clients every 10 minutes based on bandwidth if you like. No other NFS server can do this without NFS timeouts. You can loose from 1 to 4 shelves of 12 disks without data loss which is another unique feature.

Vmware currently does not support isilons smartconnect NFS failover. So you are currently stuck creating a different datastore for each node. It’s not that bad of a problem…

Setup is a snap… you can even add nodes from the front panel.

Isilon is a very young company and it shows in their sales and technical support. The don’t have many bells and whistles and their NFS and CIFS(samba) are basic, however their price reflects the offer.

The key to Netapps success is their ontap OS which supports NAS, FC and ISCSI all in one product which is amazing… Isilon’s Onefs may turnout the be they next OS of choice for storage, however it will be several years before all the features are available.

The main benefit of Isilon: It’s a wakeup call for other NAS vendors…

joe m. January 31, 2008 at 8:46 am

vo — can you give us an idea on price for the Isilon X-series?

Or at least a comparison to the netapp 3070 you benchmarked it against — Compared to a netapp 3070 with DS-14 10K FC disk shelves attached, where does the price of the Isilon setup fall? (for example) +/- 20% the cost of the netapp?

TylerB January 31, 2008 at 10:40 am

>>iometer show performance of the Xnodes under vmware are 2-5x better than the previous nodes and are inline with the iometer numbers for a Netapp 3070. So 6 nodes will perform about the same a 6 3070’s.

TylerB January 31, 2008 at 6:04 pm

My last comment was truncated-

vo- I find those performance statistics unbelievable. There is no way a single isilon node could match performance with a 3070 controller-based system, period. Not even close.

joe m. February 1, 2008 at 7:40 am

How could you (or why would you) benchmark a single Isilon node anyway? I thought you needed at least 3 nodes to form a cluster and get up and running.

If anyone has an Isilon system, can you post some benchmarking output? =) Interested in iops and bandwidth numbers.

CC March 13, 2008 at 9:02 pm

We’re using a cluster of 4 of the IQ12000X nodes with 4 of the EX12000X nodes. We are running in n+1 redundancy mode so we loose a single IQ+EX node for parity. We are mainly CIFS and do benefit from having very large file sizes (typically in the 500Mb to 1GB range.)

Performance has been very good, with read performance often hovering around 480MB/s and writes in the 330MB/s range. This level of reads is very close to wire speed (4 gig-e ports active).

It’s a very unique product that for certain workloads can’t be matched. We’ve used other NAS gateway products and they can’t compete.

When price of NAS gateway, FC fabric, and storage is compared to the Isilon pricing it is competitive.

Hope this helps….

CC

Leave a Comment

Previous post:

Next post: