June 2, 2005 at 3:32 pm
I’ve been doing a bit of research on iSCSI and Fibre SAN and here’s what I’ve found.
The King Fibre Channel SAN:
1. Speed- 4Gb is the new champion 2Gb will be the entry level.
2. Network Congestion- Since the storage is centrally located behind Fibre Switch net congestion is minimized. Server to server copies happen really fast.
3. Long Fibre cable runs.
4. Reliable proven tech
Fibre Switch costs
Fibre HBA costs
Somewhat restricted placement
Need for a RAID Array
iSCSI …potentially potent new challenger
1. Speed- Line speed which typically means GIGe but many people forget you could aggregate 4 GIGe ports at the switch for 4Gb connections. In a few years 10G will be feasible. iSCSI will have nice scalability here
2. Cost- No need for HBA unless you have a high IOP task like running an Exchange Server or DB. Freely downloadable iSCSI initiators will “trick” computer into thinking local Ethernet NIC is SCSI card.
3. Utilizes existing network infrastructure. Lower startup costs.
4. iSCSI boot. IBM booted a computer in Tel Aviv from a iSCSI Target in Seattle WA. Imagine the ease of management when you’re booting “diskless” Servers over the network.
Performance is limited to line speed
Network Congestion must be dealt with
iSCSI commands can overwhelm Server in some cases HBA helps here.
Not as proven as FC SAN but vendor support is rapidly rising.
Coolest stuff I’ve seen so far.
Alacritech has the right product mix in the SES200. You get TCP Offload for iSCSI acceleration and you keep ethernet features like the Port Trunking 802.3ad aggregation. It’s the only card I know that does this until cards start using the Broadcom BCM8708S chip which has iSCSI, TOE and RDMA on one chip.
He’s probably right. With 10G ethernet available now but very expensive, ethernet’s ubiquidity will be hard to beat.
June 9, 2005 at 1:36 pm
So what have you decided to do?
ISCSI or Firbre Channel?
June 13, 2005 at 2:18 pm
for my clients. With Gige networks becoming cheap to setup
and Fibre still being expensive I see a lot of people moving to
iSCSI and in many cases using hybrid setups where Database and Exchange servers are on FC while the rest of the network file sharing servers are iSCSI.
I’m now seeing the 1TB iSCSI RAID5 drives coming down to $3k and D-Link just announced their Xstorage (no pricing yet) but knowing D-Link it’ll be pretty damn cheap.
June 15, 2005 at 1:20 pm
One point to note through this whole discussion is the requirement, be it for FC or iSCSI based SANs, the requirement for client-side clustered file system software (assuming that the users need to share the data in the storage pool).
A NAS solution offers shared storage with the added benefit of not requiring client side software. This post is mildly self-serving as we make a NAS device, but just wanted to highlight the fact that there is an alternative to the complexity and cost of a SAN. Regards,
June 15, 2005 at 3:34 pm
Agree Michael the software management requirement of a SAN is somewhat stifling. I’ve made contact with Dataplow, they make a NAS/SAN hybrid software. Perhaps it will offer the best of both worlds. I’m checking out your products as well. Thanks.
June 23, 2005 at 5:10 pm
It doesn’t have to be complex for a smaller number of clients. Terrablock servers (Facilis Technology) don’t run a metadata ethernet system (less complexity) but still runs on Fiber Channel (4Gb or 2Gb). There are no client licenses per se but all clients share a small volume that contains the “client software”. The downside, of course, is that without a metadata server only one client can have permission to write to a volume at once. But paying $600 for a HBA is better than $1200 for client software per user.
Not the perfect solution but for smaller groups it’s inexpensive, very workable, and system administration is kept to the bare minimum (creating/deleting volumes, adding users).
Just depends on your workflow needs.
June 23, 2005 at 7:21 pm
I found one problem with a Terrablock SAN when I saw it at NAB: On the Mac clients they had to reboot the client for every time they need to change client permissions, unmount or mount new drives.
On PC’s the system seems to be pretty faultless – saw it running several AVID systems with the OMFI directory and database rebuilding problems sorted out.
August 14, 2005 at 6:21 pm
Do you have any idea how long it tok to re address/index the volumes once the reboot occured? At first I thought so what need to change a drive go around close all running programs and turn off all the Mac’s. Then I remembered that there some implications that may occur such as the index of the IO block and partition/’s. Since NLE’s use these addresses as pointers to cuts effects and keyframes would the addition and re write of volumes ruin project files?
I realize this is an old post so I am going to put it back up on top…
young gun seriously looking at the TerraBlock, or just setting up a Tiger Server with fiber connections on my own.
Log in to reply.