Isilon Systems Inc. has released an update to its OneFS software
for clustered
network attached storage (NAS) that includes
snapshots, bandwidth provisioning and improved failure
tolerance. Users and analysts said the new features are a big
step for Isilon toward giving Network Appliance Inc. (NetApp) a
run for its money in the enterprise NAS space, but that certain
features, especially backup using
NDMP, still need improvement.
Taneja Group senior analyst Brad O'Neill said the addition of
new software features, especially snapshots, are "critical" for
Isilon to make inroads into the enterprise NAS space.
"Snapshots are a really big thing. No way can they sell into the
enterprise NAS environment or the larger customer base they want to
reach without snapshots," O'Neill said.
The new SnapshotIQ allows for up to 1,024 snapshots on the
cluster, and the snapshots are performed in a way similar to
NetApp's method, which stores only changes to files and uses very
little CPU, as opposed to Hitachi Data Systems (HDS) and EMC
Corp.'s snapshot option, the Business Continuance Volume (BCV).
Meanwhile, Isilon is also differentiating its snapshots from
NetApp's in terms of the number allowed -- 1,024 at the directory
level and unlimited global snapshots; NetApp's limit is 255 per
WAFL volume. Isilon also does not require the allocation or
management of space for snapshots, since data is automatically
distributed among cluster nodes. NetApp's FlexVol virtualization
feature can perform a similar function, but its best practices
generally call for the preallocation of reserve space for snapshots
as well.
"It's great to have more granular backups than daily writes to
tape," said David Kirchhoff, manager of information technology and
senior geophysicist at Brigham Exploration Co. Previously,
Kirchhoff said he had been doing host-based snapshots on a per
server basis using Windows Volume Shadow Copy.
"That worked OK, but it was very cumbersome," Kirchhoff said. He
added that the Isilon snapshots allowed him to set snapshot
policies and schedules down to the file level, according him more
flexibility than Shadow Copy in terms of which particular files he
backs up and how often.
Moving toward a new customer base
Meanwhile, at least a few users not from Isilon's traditional
high-performance computing playground have begun to pick up the
system, including the
county of San Bernardino, Calif., and
broadcasting giant Clear Channel Communications.
According to Greg Robinson, solution architect for Clear Channel
Communications, the company uses a 24-node, 35 terabyte (TB) Isilon
cluster to store data from a farm of servers that host radio
station Web sites. Robinson said the company also uses the cluster
for files from EMC's EmailXtender email archiving software using
SyncIQ, Isilon's replication software, and it is looking into
adding a cluster in Denver for disaster recovery of corporate
data.
Robinson said he had previously used EMC's Celerra NAS systems
but had had difficulty upgrading capacity and experienced issues
with file system scalability. He admits he's not the stereotypical
Isilon user, but that the ability to add storage and throughput to
the Isilon system won him over. Clear Channel is still using EMC
for its primary
storage area network (SAN) storage and plans
to keep it that way.
"Even if Isilon added block-level access, in the form of iSCSI,
I probably still wouldn't use it," Robinson said. "I don't think
iSCSI is really an enterprise play, and we wouldn't have a use for
it."
Robinson also said he missed one aspect of the Celerra clusters
he had been using -- ease of use with NDMP. With the Celerra
clusters, metadata could be transferred quickly to his Legato
backup server over IP, while the backup data itself could be sent
speedily through Fibre Channel (FC). Currently, Robinson said, the
Isilon system has
InfiniBand interconnects between nodes, but
it still uses Gigabit Ethernet (GigE) out the back end, making
NDMP for backups "not worth it" and not something he's using
after trying it for a short time around a year ago.
"I would like to extend that InfiniBand performance to clients,
namely my backup client," Robinson said. "I can still do my backups
to tape from the cluster, but I could get better performance out of
it."
Kirchhoff said he had been having problems getting NDMP to work
at all with his Legato backup server.
"Legato supports NDMP but has not certified it with Isilon,"
Kirchhoff said. "They just need to get together and get the
protocols and versions [right] so it will work better."
Currently, according to Kirchhoff, Brigham is backing up from a
Network File System (NFS) mount on a separate Solaris 8 server. "It
works fine, but NDMP would probably be faster," he said. "I'm not
losing any functionality for backups, just some speed." That speed
will become more important next year, when a new 3D land survey
over 180 square miles of northern Texas and southern Oklahoma gets
integrated into the San Antonio-based oil and gas exploration
firm's storage system. The extremely dense data associated with
that study, Kirchhoff said, will probably mean his Isilon cluster
will have to at least double in size -- as will his backups.
Reached for comment with regard to the NDMP issue, Isilon
spokesman Lucas Welch confirmed that Isilon does not yet have
certification with Legato, but that it was in the process of
qualification with both Legato and CommVault, and it already has
been certified with Symantec Corp.'s NetBackup. "Isilon's customers
have multiple ways of backing up and restoring their data," Welch
added. "NDMP is just one option among several."
"Isilon's not quite ready yet for general purpose NAS,"
according to Tony Asaro, analyst for the Enterprise Strategy Group.
"The core of the market is still not using them -- engineering
development departments may have a home directory on Isilon, but
the rest of corporate America isn't going to store their
mission-critical data on their system."
Taneja Group's O'Neill added that if Isilon is casting an eye
toward the enterprise NAS space, its technology should ideally also
support block-level access as NetApp's NAS/SAN systems do.
Archiving and compliance features would also be a plus, he
said.
"Their scalability and performance story is awesome," O'Neill
said. "They've already been able to make hay against NetApp in the
high-performance computing market, and the work they've already
done with NAS is not trivial."
Specifications on the new software
OneFS version 4.5 will now allow Isilon's IQ clusters to scale
up to 96 nodes, with one petabyte possible under a single file
system. According to Isilon founder and chief technology officer
Sujal Patel, previously the biggest clusters in actual
implementation belonged to MySpace.com, at 53 nodes. But, he said,
many of Isilon's customers, who traditionally have been at the very
bleeding edge of performance and capacity requirements, have
surpassed cluster limits every time they've been raised.
With aggregate throughput, the new capacity ceiling for clusters
means that the maximum performance for the IQ system has been
raised to 10 gigabytes per second (GBps). Isilon has added greater
failure tolerance to the system; now, up to three or four nodes
(depending on a user's licensing of the software) can fail without
downtime. Patel said the new failover design had been designed with
the same storage overhead -- some 20% to 25 % of capacity -- as
one- or two-node failover designs.
Lastly, the new SmartConnect feature allows users to zone
bandwidth to portions of the cluster and to direct different
traffic streams to different nodes of the cluster by virtualizing
their IP addresses.
"This creates a DNS [domain name system] delegation zone and
balances all the IPs by throughput," Kirchhoff said. "That's good,
because it means that all the traffic isn't coming in on one
specific IP address or node of the cluster."
The addition of the feature also quells some competitors'
attempts to muddy the waters around Isilon. Clustered NAS newcomer
Crosswalk Inc.
came out swinging against Isilon this past
April, calling its method of breaking up the file system through
one node a bottleneck.
SmartConnect is available starting at $4,950 per node, and
SnapshotIQ is available starting at $2,950 per node.