Re: deletion performance of large deduplicated files
Appreciate the input Del, thanks ! Thank you, -Rick Adamson -Original Message- From: ADSM: Dist Stor Manager On Behalf Of Del Hoobler Sent: Friday, July 19, 2019 2:32 PM To: ADSM-L@VM.MARIST.EDU Subject: Re: [ADSM-L] deletion performance of large deduplicated files * This email originated outside of the organization. Use caution when opening attachments or clicking links. * -- Hi Eric and all, We are aware of this situation and putting additional focus and attention on it. Del "ADSM: Dist Stor Manager" wrote on 07/19/2019 09:35:37 AM: > From: "Loon, Eric van (ITOP NS) - KLM" > To: ADSM-L@VM.MARIST.EDU > Date: 07/19/2019 09:36 AM > Subject: [EXTERNAL] Re: deletion performance of large deduplicated files > Sent by: "ADSM: Dist Stor Manager" > > Hi Rick and others! > > I replicated the data of the test TDP client to multiple servers, > running 7.1.7, 7.1.9 and even 8.1.8: the performance sucks on all servers. > We do not use client replication as part of our server protection. > We need a real time replication over the datacenter and thus we rely > on host-based replication for all our servers. I tested with and > without replication: there is no noticeable difference. > As a work-around I will install a non-deduplicated file stgpool on > the worst performing server next week. I expect the server to > perform better as soon as all container pool objects are expired, in > about 3 weeks from now. > In the meantime I will keep on pursuing IBM until it is fixed or > else we might need to replace the product altogether... > > Kind regards, > Eric van Loon > Air France/KLM Storage & Backup > > -Original Message- > From: ADSM: Dist Stor Manager [mailto:ADSM-L@VM.MARIST.EDU] On > Behalf Of Rick Adamson > Sent: vrijdag 19 juli 2019 14:45 > To: ADSM-L@VM.MARIST.EDU > Subject: Re: deletion performance of large deduplicated files > > Eric, Michael, > I have been working through similar struggles and as I read your > posts had to wonder, can you provide some details on your server and > client versions? > Basically I now have experience/exposure to every version/ > maintenance pack/patch SP has put out since 7.1.3.x > > My servers are now running on 8.1.8 (windows) and has seemed to > have stabilized (for now). > One thing that was causing me a lot of grief was using directory > storage pools with client side deduplication enabled, particularly > on data protection products (all of them). > Afterwards there was a lot of cleanup; auditing containers, > confirming that protect storage was completing successfully, and > finally doing the same for replicate node processes. > > I found that understandably server performance takes a severe nose > dive if you are trying to process (protect/replicate) damaged > containers, and most likely restores will be compromised as well. > > > Thank you, > -Rick Adamson > > > -Original Message- > From: ADSM: Dist Stor Manager On Behalf Of Michael Prix > Sent: Friday, July 19, 2019 6:11 AM > To: ADSM-L@VM.MARIST.EDU > Subject: Re: [ADSM-L] deletion performance of large deduplicated files > > * This email originated outside of the organization. Use caution > when opening attachments or clicking links. * > > -- > Hello Eric, > > > > welcome to my nightmares. Take a seat, wanna have a drink? > > > > I had the pleasure of performance and data corruption PMRs during the last two > > years with TDP Oracle. Yes, at first the customer got blamed for not adhering > > completely to to blueprints, but after some weeks it boild down to ... > > silence. > > Data corruption was because of what ended in IT28096 - now fixed. > > Performance is interesting, but resembles to what you have written. We work > > with MAXPIECESIZE settings on RMAN to keep the backup pieces small and got > > some interesting values, pending further observation, but we might be on a > > cheerful way. I'm talking about database sizes of 50TB here, warehouse style. > > In between we moved the big DBs to a dedicated server to prove that the > > performance drop is because of the big DBs, and the remaining "small" DBs - > > size of 500MB up to 5TB - didn't put any measurable stress on the DB in terms > > of expiration and protect stgpool. Even the big DBs on their dedicated server > > performed better in terms of expiration and protect stgpool, which might have > > been a coincidence of these DBs holding nearly the same data and having the > > same retention period. > > > > What I can't observe is a slowness of the DB. Queries are answered in the > > normal time - depending on the query. a count(*) from backupobjects naturally > > takes some time, considerably longer when you use dedup, but the daily queries > > are answered in the "normal" timeframe. > > > > What helped immediately was some tuning: > >
Re: deletion performance of large deduplicated files
Hi Eric and all, We are aware of this situation and putting additional focus and attention on it. Del "ADSM: Dist Stor Manager" wrote on 07/19/2019 09:35:37 AM: > From: "Loon, Eric van (ITOP NS) - KLM" > To: ADSM-L@VM.MARIST.EDU > Date: 07/19/2019 09:36 AM > Subject: [EXTERNAL] Re: deletion performance of large deduplicated files > Sent by: "ADSM: Dist Stor Manager" > > Hi Rick and others! > > I replicated the data of the test TDP client to multiple servers, > running 7.1.7, 7.1.9 and even 8.1.8: the performance sucks on all servers. > We do not use client replication as part of our server protection. > We need a real time replication over the datacenter and thus we rely > on host-based replication for all our servers. I tested with and > without replication: there is no noticeable difference. > As a work-around I will install a non-deduplicated file stgpool on > the worst performing server next week. I expect the server to > perform better as soon as all container pool objects are expired, in > about 3 weeks from now. > In the meantime I will keep on pursuing IBM until it is fixed or > else we might need to replace the product altogether... > > Kind regards, > Eric van Loon > Air France/KLM Storage & Backup > > -Original Message- > From: ADSM: Dist Stor Manager [mailto:ADSM-L@VM.MARIST.EDU] On > Behalf Of Rick Adamson > Sent: vrijdag 19 juli 2019 14:45 > To: ADSM-L@VM.MARIST.EDU > Subject: Re: deletion performance of large deduplicated files > > Eric, Michael, > I have been working through similar struggles and as I read your > posts had to wonder, can you provide some details on your server and > client versions? > Basically I now have experience/exposure to every version/ > maintenance pack/patch SP has put out since 7.1.3.x > > My servers are now running on 8.1.8 (windows) and has seemed to > have stabilized (for now). > One thing that was causing me a lot of grief was using directory > storage pools with client side deduplication enabled, particularly > on data protection products (all of them). > Afterwards there was a lot of cleanup; auditing containers, > confirming that protect storage was completing successfully, and > finally doing the same for replicate node processes. > > I found that understandably server performance takes a severe nose > dive if you are trying to process (protect/replicate) damaged > containers, and most likely restores will be compromised as well. > > > Thank you, > -Rick Adamson > > > -Original Message- > From: ADSM: Dist Stor Manager On Behalf Of Michael Prix > Sent: Friday, July 19, 2019 6:11 AM > To: ADSM-L@VM.MARIST.EDU > Subject: Re: [ADSM-L] deletion performance of large deduplicated files > > * This email originated outside of the organization. Use caution > when opening attachments or clicking links. * > > -- > Hello Eric, > > > > welcome to my nightmares. Take a seat, wanna have a drink? > > > > I had the pleasure of performance and data corruption PMRs during the last two > > years with TDP Oracle. Yes, at first the customer got blamed for not adhering > > completely to to blueprints, but after some weeks it boild down to ... > > silence. > > Data corruption was because of what ended in IT28096 - now fixed. > > Performance is interesting, but resembles to what you have written. We work > > with MAXPIECESIZE settings on RMAN to keep the backup pieces small and got > > some interesting values, pending further observation, but we might be on a > > cheerful way. I'm talking about database sizes of 50TB here, warehouse style. > > In between we moved the big DBs to a dedicated server to prove that the > > performance drop is because of the big DBs, and the remaining "small" DBs - > > size of 500MB up to 5TB - didn't put any measurable stress on the DB in terms > > of expiration and protect stgpool. Even the big DBs on their dedicated server > > performed better in terms of expiration and protect stgpool, which might have > > been a coincidence of these DBs holding nearly the same data and having the > > same retention period. > > > > What I can't observe is a slowness of the DB. Queries are answered in the > > normal time - depending on the query. a count(*) from backupobjects naturally > > takes some time, considerably longer when you use dedup, but the daily queries > > are answered in the "normal" timeframe. > > > > What helped immediately was some tuning: > > - More LUNS and filesystems for the TSM-DB > > - smaller disks, but more of them, for each filesystem. > > changing the disks from 100GB to 2 x 50GB for each DB-filesystem got me a > > performance boost of 200% in expiration and backup db. Unbelievable,but true. > > Yes, I'm using SSD. And SVC. And multiple storage systems. Performance isn't > > the problem, we are
Re: deletion performance of large deduplicated files
OK. Thanks. Michael. We'll give it a try. Chris Kizzire Backup Administrator (Network Engineer II) BROOKWOOD BAPTIST HEALTH Information Systems O: 205.820.5973 chris.kizz...@bhsala.com BROOKWOODBAPTISTHEALTH.COM -Original Message- From: ADSM: Dist Stor Manager On Behalf Of Michael Prix Sent: Friday, July 19, 2019 10:14 AM To: ADSM-L@VM.MARIST.EDU Subject: Re: [ADSM-L] deletion performance of large deduplicated files CAUTION: ***EXTERNAL EMAIL*** Do NOT click links or open attachments unless you recognize the sender and know the content is safe. If you are unsure, please use PhishAlarm to report suspicious emails. Chris, yours is easy to answer: You have a performance problem, not a TSM problem. For VM-backup I have a dedicated LPAR in a S842, 3vCPU, 128GB RAM, SR-IOV 10GB. Storage is a V7000, SSD/SAS. There has never been a backup or restore problem performance wise, everything is running with wire-speed. The only problem we hsd was client side dedup, II28096 hit us bad, 5 days of audit for each TSM-instance over and over again until the reason was identified. For a start: keep the size of hdisk small for the db. Better 2 x 50GB that 1 100GB. chfs -e x and reorgvg is your friend. mount option "rbrw,noatime" is of great use for db, log, arc and stg pools. Check you queue_depth and set it to max. As for SQL: use as many streams as the client can handle. CPU is there to handle load, not to waste energy by idleing. -- Michael Prix On Fri, 2019-07-19 at 13:02 +, Kizzire, Chris wrote: > Alas... we are not the only ones I knew it... > We went from TSM 6.3 to SP 8.1.4.0. Performance is 5 times slower in > the new Environment overall. We use Container Pools, Dedup, & > Compression. We use SP for VE for most VM's. Baclient & SQL for physical > machines & vm's w/ SQL. > It took about 17 hours to restore 4.5TB SQL DB the other day. > Main Server: > IBM Power 750 running AIX 7.2 . > 3.1TB DB is on SSD. > 128GB RAM > Server at DR site > IBM Power 770 running AIX 7.2 > 3.1TB DB NOT on SSD > 100ish GB RAM > Container Pool is on 1.5 year old IBM V5030 w/ Identical V5030 at DR > site. > > IBM says open a Performance PMR for AIX- which we have yet to do. > Protect Stgpool runs for days & we have to cancel because we get too > far behind on Replication. If we are lucky we might can replicate 12TB > in a 24 hour period w/ 100 sessions (maxsessions=50) > > > Chris Kizzire > Backup Administrator (Network Engineer II) > > BROOKWOOD BAPTIST HEALTH > Information Systems > O: 205.820.5973 > > chris.kizz...@bhsala.com > BROOKWOODBAPTISTHEALTH.COM > > > -Original Message- > From: ADSM: Dist Stor Manager On Behalf Of > Michael Prix > Sent: Friday, July 19, 2019 5:11 AM > To: ADSM-L@VM.MARIST.EDU > Subject: Re: [ADSM-L] deletion performance of large deduplicated files > > > CAUTION: ***EXTERNAL EMAIL*** Do NOT click links or open attachments > unless you recognize the sender and know the content is safe. If you > are unsure, please use PhishAlarm to report suspicious emails. > > Hello Eric, > > welcome to my nightmares. Take a seat, wanna have a drink? > > I had the pleasure of performance and data corruption PMRs during the > last two years with TDP Oracle. Yes, at first the customer got blamed > for not adhering completely to to blueprints, but after some weeks it > boild down to ... > silence. > Data corruption was because of what ended in IT28096 - now fixed. > Performance is interesting, but resembles to what you have written. We > work with MAXPIECESIZE settings on RMAN to keep the backup pieces > small and got some interesting values, pending further observation, > but we might be on a cheerful way. I'm talking about database sizes of > 50TB here, warehouse style. > In between we moved the big DBs to a dedicated server to prove that > the performance drop is because of the big DBs, and the remaining > "small" DBs - size of 500MB up to 5TB - didn't put any measurable > stress on the DB in terms of expiration and protect stgpool. Even the > big DBs on their dedicated server performed better in terms of > expiration and protect stgpool, which might have been a coincidence of > these DBs holding nearly the same data and having the same retention period. > > What I can't observe is a slowness of the DB. Queries are answered in > the normal time - depending on the query. a count(*) from > backupobjects naturally takes some time, considerably longer when you > use dedup, but the daily queries are answered in the "normal" timeframe. > > What helped immediately was some tuning: > - More LUNS and filesystems for the TSM-DB > - smaller disks, but more of them, for each filesystem. > changing the disks from 100GB to 2 x 50GB for each DB-filesystem got > me a performance boost of 200% in expiration and backup db. > Unbelievable, but true. > Yes, I'm using SSD. And SVC. And multiple storage systems. Performance >
Re: deletion performance of large deduplicated files
Hello Eric, could you describe a bit you server setup? What type of data are you storing? SQL (which), VM, File, ... Is everything going into one dedup pool or are they split for different types of data? I assume your TSM-servers run on AIX. In general, performant backup and restore is no problem with deduplicated storagepools, if done the right way. Housekeeping is a total different pair of shoes. -- Michael Prix On Fri, 2019-07-19 at 13:35 +, Loon, Eric van (ITOP NS) - KLM wrote: > Hi Rick and others! > > I replicated the data of the test TDP client to multiple servers, running > 7.1.7, 7.1.9 and even 8.1.8: the performance sucks on all servers. > We do not use client replication as part of our server protection. We need a > real time replication over the datacenter and thus we rely on host-based > replication for all our servers. I tested with and without replication: > there is no noticeable difference. > As a work-around I will install a non-deduplicated file stgpool on the worst > performing server next week. I expect the server to perform better as soon > as all container pool objects are expired, in about 3 weeks from now. > In the meantime I will keep on pursuing IBM until it is fixed or else we > might need to replace the product altogether... > > Kind regards, > Eric van Loon > Air France/KLM Storage & Backup > > -Original Message- > From: ADSM: Dist Stor Manager [mailto:ADSM-L@VM.MARIST.EDU] On Behalf Of > Rick Adamson > Sent: vrijdag 19 juli 2019 14:45 > To: ADSM-L@VM.MARIST.EDU > Subject: Re: deletion performance of large deduplicated files > > Eric, Michael, > I have been working through similar struggles and as I read your posts had > to wonder, can you provide some details on your server and client versions? > Basically I now have experience/exposure to every version/maintenance > pack/patch SP has put out since 7.1.3.x > > My servers are now running on 8.1.8 (windows) and has seemed to have > stabilized (for now). > One thing that was causing me a lot of grief was using directory storage > pools with client side deduplication enabled, particularly on data > protection products (all of them). > Afterwards there was a lot of cleanup; auditing containers, confirming that > protect storage was completing successfully, and finally doing the same for > replicate node processes. > > I found that understandably server performance takes a severe nose dive if > you are trying to process (protect/replicate) damaged containers, and most > likely restores will be compromised as well. > > > Thank you, > -Rick Adamson > > > -Original Message- > From: ADSM: Dist Stor Manager On Behalf Of Michael > Prix > Sent: Friday, July 19, 2019 6:11 AM > To: ADSM-L@VM.MARIST.EDU > Subject: Re: [ADSM-L] deletion performance of large deduplicated files > > * This email originated outside of the organization. Use caution when > opening attachments or clicking links. * > > -- > Hello Eric, > > > > welcome to my nightmares. Take a seat, wanna have a drink? > > > > I had the pleasure of performance and data corruption PMRs during the last > two > > years with TDP Oracle. Yes, at first the customer got blamed for not > adhering > > completely to to blueprints, but after some weeks it boild down to ... > > silence. > > Data corruption was because of what ended in IT28096 - now fixed. > > Performance is interesting, but resembles to what you have written. We work > > with MAXPIECESIZE settings on RMAN to keep the backup pieces small and got > > some interesting values, pending further observation, but we might be on a > > cheerful way. I'm talking about database sizes of 50TB here, warehouse > style. > > In between we moved the big DBs to a dedicated server to prove that the > > performance drop is because of the big DBs, and the remaining "small" DBs - > > size of 500MB up to 5TB - didn't put any measurable stress on the DB in > terms > > of expiration and protect stgpool. Even the big DBs on their dedicated > server > > performed better in terms of expiration and protect stgpool, which might > have > > been a coincidence of these DBs holding nearly the same data and having the > > same retention period. > > > > What I can't observe is a slowness of the DB. Queries are answered in the > > normal time - depending on the query. a count(*) from backupobjects > naturally > > takes some time, considerably longer when you use dedup, but the daily > queries > > are answered in the "normal" timeframe. > > > > What helped immediately was some tuning: > > - More LUNS and filesystems for the TSM-DB > > - smaller disks, but more of them, for each filesystem. > > changing the disks from 100GB to 2 x 50GB for each DB-filesystem got me a > > performance boost of 200% in expiration and backup db. Unbelievable, but > true. > > Yes, I'm using SSD. And SVC. And multiple storage systems. Performance isn't > > the problem, we are measuring
Re: deletion performance of large deduplicated files
Chris, yours is easy to answer: You have a performance problem, not a TSM problem. For VM-backup I have a dedicated LPAR in a S842, 3vCPU, 128GB RAM, SR-IOV 10GB. Storage is a V7000, SSD/SAS. There has never been a backup or restore problem performance wise, everything is running with wire-speed. The only problem we hsd was client side dedup, II28096 hit us bad, 5 days of audit for each TSM-instance over and over again until the reason was identified. For a start: keep the size of hdisk small for the db. Better 2 x 50GB that 1 100GB. chfs -e x and reorgvg is your friend. mount option "rbrw,noatime" is of great use for db, log, arc and stg pools. Check you queue_depth and set it to max. As for SQL: use as many streams as the client can handle. CPU is there to handle load, not to waste energy by idleing. -- Michael Prix On Fri, 2019-07-19 at 13:02 +, Kizzire, Chris wrote: > Alas... we are not the only ones I knew it... > We went from TSM 6.3 to SP 8.1.4.0. Performance is 5 times slower in the new > Environment overall. We use Container Pools, Dedup, & Compression. We use SP > for VE for most VM's. Baclient & SQL for physical machines & vm's w/ SQL. > It took about 17 hours to restore 4.5TB SQL DB the other day. > Main Server: > IBM Power 750 running AIX 7.2 . > 3.1TB DB is on SSD. > 128GB RAM > Server at DR site > IBM Power 770 running AIX 7.2 > 3.1TB DB NOT on SSD > 100ish GB RAM > Container Pool is on 1.5 year old IBM V5030 > w/ Identical V5030 at DR site. > > IBM says open a Performance PMR for AIX- which we have yet to do. > Protect Stgpool runs for days & we have to cancel because we get too far > behind on Replication. If we are lucky we might can replicate 12TB in a 24 > hour period w/ 100 sessions (maxsessions=50) > > > Chris Kizzire > Backup Administrator (Network Engineer II) > > BROOKWOOD BAPTIST HEALTH > Information Systems > O: 205.820.5973 > > chris.kizz...@bhsala.com > BROOKWOODBAPTISTHEALTH.COM > > > -Original Message- > From: ADSM: Dist Stor Manager On Behalf Of Michael > Prix > Sent: Friday, July 19, 2019 5:11 AM > To: ADSM-L@VM.MARIST.EDU > Subject: Re: [ADSM-L] deletion performance of large deduplicated files > > > CAUTION: ***EXTERNAL EMAIL*** Do NOT click links or open attachments unless > you recognize the sender and know the content is safe. If you are unsure, > please use PhishAlarm to report suspicious emails. > > Hello Eric, > > welcome to my nightmares. Take a seat, wanna have a drink? > > I had the pleasure of performance and data corruption PMRs during the last > two years with TDP Oracle. Yes, at first the customer got blamed for not > adhering completely to to blueprints, but after some weeks it boild down > to ... > silence. > Data corruption was because of what ended in IT28096 - now fixed. > Performance is interesting, but resembles to what you have written. We work > with MAXPIECESIZE settings on RMAN to keep the backup pieces small and got > some interesting values, pending further observation, but we might be on a > cheerful way. I'm talking about database sizes of 50TB here, warehouse > style. > In between we moved the big DBs to a dedicated server to prove that the > performance drop is because of the big DBs, and the remaining "small" DBs - > size of 500MB up to 5TB - didn't put any measurable stress on the DB in > terms of expiration and protect stgpool. Even the big DBs on their dedicated > server performed better in terms of expiration and protect stgpool, which > might have been a coincidence of these DBs holding nearly the same data and > having the same retention period. > > What I can't observe is a slowness of the DB. Queries are answered in the > normal time - depending on the query. a count(*) from backupobjects > naturally takes some time, considerably longer when you use dedup, but the > daily queries are answered in the "normal" timeframe. > > What helped immediately was some tuning: > - More LUNS and filesystems for the TSM-DB > - smaller disks, but more of them, for each filesystem. > changing the disks from 100GB to 2 x 50GB for each DB-filesystem got me a > performance boost of 200% in expiration and backup db. Unbelievable, but > true. > Yes, I'm using SSD. And SVC. And multiple storage systems. Performance isn't > the problem, we are measuring 2ms respone time for write AND read. > - stripeset for each fileset > > > -- > Michael Prix > > On Fri, 2019-07-19 at 07:29 +, Loon, Eric van (ITOP NS) - KLM wrote: > > Hi TSM/SP-ers, > > > > We are struggling with the performance of our TSM servers for months > > now. We are running several servers with hardware (Data Domain) dedup > > for years without any problems, but on our new servers with directory > > container pools performance is really, really bad. > > The servers and storage are designed according to the Blueprints and > > they are working fine as long as you do not add large database (Oracle > > and SAP) client to them. As
Re: deletion performance of large deduplicated files
Hi Rick and others! I replicated the data of the test TDP client to multiple servers, running 7.1.7, 7.1.9 and even 8.1.8: the performance sucks on all servers. We do not use client replication as part of our server protection. We need a real time replication over the datacenter and thus we rely on host-based replication for all our servers. I tested with and without replication: there is no noticeable difference. As a work-around I will install a non-deduplicated file stgpool on the worst performing server next week. I expect the server to perform better as soon as all container pool objects are expired, in about 3 weeks from now. In the meantime I will keep on pursuing IBM until it is fixed or else we might need to replace the product altogether... Kind regards, Eric van Loon Air France/KLM Storage & Backup -Original Message- From: ADSM: Dist Stor Manager [mailto:ADSM-L@VM.MARIST.EDU] On Behalf Of Rick Adamson Sent: vrijdag 19 juli 2019 14:45 To: ADSM-L@VM.MARIST.EDU Subject: Re: deletion performance of large deduplicated files Eric, Michael, I have been working through similar struggles and as I read your posts had to wonder, can you provide some details on your server and client versions? Basically I now have experience/exposure to every version/maintenance pack/patch SP has put out since 7.1.3.x My servers are now running on 8.1.8 (windows) and has seemed to have stabilized (for now). One thing that was causing me a lot of grief was using directory storage pools with client side deduplication enabled, particularly on data protection products (all of them). Afterwards there was a lot of cleanup; auditing containers, confirming that protect storage was completing successfully, and finally doing the same for replicate node processes. I found that understandably server performance takes a severe nose dive if you are trying to process (protect/replicate) damaged containers, and most likely restores will be compromised as well. Thank you, -Rick Adamson -Original Message- From: ADSM: Dist Stor Manager On Behalf Of Michael Prix Sent: Friday, July 19, 2019 6:11 AM To: ADSM-L@VM.MARIST.EDU Subject: Re: [ADSM-L] deletion performance of large deduplicated files * This email originated outside of the organization. Use caution when opening attachments or clicking links. * -- Hello Eric, welcome to my nightmares. Take a seat, wanna have a drink? I had the pleasure of performance and data corruption PMRs during the last two years with TDP Oracle. Yes, at first the customer got blamed for not adhering completely to to blueprints, but after some weeks it boild down to ... silence. Data corruption was because of what ended in IT28096 - now fixed. Performance is interesting, but resembles to what you have written. We work with MAXPIECESIZE settings on RMAN to keep the backup pieces small and got some interesting values, pending further observation, but we might be on a cheerful way. I'm talking about database sizes of 50TB here, warehouse style. In between we moved the big DBs to a dedicated server to prove that the performance drop is because of the big DBs, and the remaining "small" DBs - size of 500MB up to 5TB - didn't put any measurable stress on the DB in terms of expiration and protect stgpool. Even the big DBs on their dedicated server performed better in terms of expiration and protect stgpool, which might have been a coincidence of these DBs holding nearly the same data and having the same retention period. What I can't observe is a slowness of the DB. Queries are answered in the normal time - depending on the query. a count(*) from backupobjects naturally takes some time, considerably longer when you use dedup, but the daily queries are answered in the "normal" timeframe. What helped immediately was some tuning: - More LUNS and filesystems for the TSM-DB - smaller disks, but more of them, for each filesystem. changing the disks from 100GB to 2 x 50GB for each DB-filesystem got me a performance boost of 200% in expiration and backup db. Unbelievable, but true. Yes, I'm using SSD. And SVC. And multiple storage systems. Performance isn't the problem, we are measuring 2ms respone time for write AND read. - stripeset for each fileset -- Michael Prix On Fri, 2019-07-19 at 07:29 +, Loon, Eric van (ITOP NS) - KLM wrote: > Hi TSM/SP-ers, > > We are struggling with the performance of our TSM servers for months now. We > are running several servers with hardware (Data Domain) dedup for years > without any problems, but on our new servers with directory container pools > performance is really, really bad. > The servers and storage are designed according to the Blueprints and they > are working fine as long as you do not add large database (Oracle and SAP) > client to them. As soon as you do, the overall server performance becomes > very
Re: deletion performance of large deduplicated files
Alas... we are not the only ones I knew it... We went from TSM 6.3 to SP 8.1.4.0. Performance is 5 times slower in the new Environment overall. We use Container Pools, Dedup, & Compression. We use SP for VE for most VM's. Baclient & SQL for physical machines & vm's w/ SQL. It took about 17 hours to restore 4.5TB SQL DB the other day. Main Server: IBM Power 750 running AIX 7.2 . 3.1TB DB is on SSD. 128GB RAM Server at DR site IBM Power 770 running AIX 7.2 3.1TB DB NOT on SSD 100ish GB RAM Container Pool is on 1.5 year old IBM V5030 w/ Identical V5030 at DR site. IBM says open a Performance PMR for AIX- which we have yet to do. Protect Stgpool runs for days & we have to cancel because we get too far behind on Replication. If we are lucky we might can replicate 12TB in a 24 hour period w/ 100 sessions (maxsessions=50) Chris Kizzire Backup Administrator (Network Engineer II) BROOKWOOD BAPTIST HEALTH Information Systems O: 205.820.5973 chris.kizz...@bhsala.com BROOKWOODBAPTISTHEALTH.COM -Original Message- From: ADSM: Dist Stor Manager On Behalf Of Michael Prix Sent: Friday, July 19, 2019 5:11 AM To: ADSM-L@VM.MARIST.EDU Subject: Re: [ADSM-L] deletion performance of large deduplicated files CAUTION: ***EXTERNAL EMAIL*** Do NOT click links or open attachments unless you recognize the sender and know the content is safe. If you are unsure, please use PhishAlarm to report suspicious emails. Hello Eric, welcome to my nightmares. Take a seat, wanna have a drink? I had the pleasure of performance and data corruption PMRs during the last two years with TDP Oracle. Yes, at first the customer got blamed for not adhering completely to to blueprints, but after some weeks it boild down to ... silence. Data corruption was because of what ended in IT28096 - now fixed. Performance is interesting, but resembles to what you have written. We work with MAXPIECESIZE settings on RMAN to keep the backup pieces small and got some interesting values, pending further observation, but we might be on a cheerful way. I'm talking about database sizes of 50TB here, warehouse style. In between we moved the big DBs to a dedicated server to prove that the performance drop is because of the big DBs, and the remaining "small" DBs - size of 500MB up to 5TB - didn't put any measurable stress on the DB in terms of expiration and protect stgpool. Even the big DBs on their dedicated server performed better in terms of expiration and protect stgpool, which might have been a coincidence of these DBs holding nearly the same data and having the same retention period. What I can't observe is a slowness of the DB. Queries are answered in the normal time - depending on the query. a count(*) from backupobjects naturally takes some time, considerably longer when you use dedup, but the daily queries are answered in the "normal" timeframe. What helped immediately was some tuning: - More LUNS and filesystems for the TSM-DB - smaller disks, but more of them, for each filesystem. changing the disks from 100GB to 2 x 50GB for each DB-filesystem got me a performance boost of 200% in expiration and backup db. Unbelievable, but true. Yes, I'm using SSD. And SVC. And multiple storage systems. Performance isn't the problem, we are measuring 2ms respone time for write AND read. - stripeset for each fileset -- Michael Prix On Fri, 2019-07-19 at 07:29 +, Loon, Eric van (ITOP NS) - KLM wrote: > Hi TSM/SP-ers, > > We are struggling with the performance of our TSM servers for months > now. We are running several servers with hardware (Data Domain) dedup > for years without any problems, but on our new servers with directory > container pools performance is really, really bad. > The servers and storage are designed according to the Blueprints and > they are working fine as long as you do not add large database (Oracle > and SAP) client to them. As soon as you do, the overall server > performance becomes very bad: client and admin session initiation > takes 20 to 40 seconds, SQL queries run for minutes where they should > take a few seconds and q query stgpool sometimes takes more than a minute to > respond! > I have two cases open for this. In one case we focused a lot on the OS > and disk performance, but during that process I noticed that the > performance is most likely caused by the way TSM processes large (few > hundred MB) files. I performed a large amount of tests and came to the > conclusion that it takes TSM a huge amount of time to delete large > deduplicated files, both in container pools as deduplicated file > pools. As test I use an TDP for Oracle client which uses a backup > piece size of 900 MB. The client contains about > 5000 files. Deleting the files from a container pool takes more than > an hour. When you run a delete object for the files individually I see > that most files take more than a second(!) to delete. If I put that
Re: deletion performance of large deduplicated files
Eric, Michael, I have been working through similar struggles and as I read your posts had to wonder, can you provide some details on your server and client versions? Basically I now have experience/exposure to every version/maintenance pack/patch SP has put out since 7.1.3.x My servers are now running on 8.1.8 (windows) and has seemed to have stabilized (for now). One thing that was causing me a lot of grief was using directory storage pools with client side deduplication enabled, particularly on data protection products (all of them). Afterwards there was a lot of cleanup; auditing containers, confirming that protect storage was completing successfully, and finally doing the same for replicate node processes. I found that understandably server performance takes a severe nose dive if you are trying to process (protect/replicate) damaged containers, and most likely restores will be compromised as well. Thank you, -Rick Adamson -Original Message- From: ADSM: Dist Stor Manager On Behalf Of Michael Prix Sent: Friday, July 19, 2019 6:11 AM To: ADSM-L@VM.MARIST.EDU Subject: Re: [ADSM-L] deletion performance of large deduplicated files * This email originated outside of the organization. Use caution when opening attachments or clicking links. * -- Hello Eric, welcome to my nightmares. Take a seat, wanna have a drink? I had the pleasure of performance and data corruption PMRs during the last two years with TDP Oracle. Yes, at first the customer got blamed for not adhering completely to to blueprints, but after some weeks it boild down to ... silence. Data corruption was because of what ended in IT28096 - now fixed. Performance is interesting, but resembles to what you have written. We work with MAXPIECESIZE settings on RMAN to keep the backup pieces small and got some interesting values, pending further observation, but we might be on a cheerful way. I'm talking about database sizes of 50TB here, warehouse style. In between we moved the big DBs to a dedicated server to prove that the performance drop is because of the big DBs, and the remaining "small" DBs - size of 500MB up to 5TB - didn't put any measurable stress on the DB in terms of expiration and protect stgpool. Even the big DBs on their dedicated server performed better in terms of expiration and protect stgpool, which might have been a coincidence of these DBs holding nearly the same data and having the same retention period. What I can't observe is a slowness of the DB. Queries are answered in the normal time - depending on the query. a count(*) from backupobjects naturally takes some time, considerably longer when you use dedup, but the daily queries are answered in the "normal" timeframe. What helped immediately was some tuning: - More LUNS and filesystems for the TSM-DB - smaller disks, but more of them, for each filesystem. changing the disks from 100GB to 2 x 50GB for each DB-filesystem got me a performance boost of 200% in expiration and backup db. Unbelievable, but true. Yes, I'm using SSD. And SVC. And multiple storage systems. Performance isn't the problem, we are measuring 2ms respone time for write AND read. - stripeset for each fileset -- Michael Prix On Fri, 2019-07-19 at 07:29 +, Loon, Eric van (ITOP NS) - KLM wrote: > Hi TSM/SP-ers, > > We are struggling with the performance of our TSM servers for months now. We > are running several servers with hardware (Data Domain) dedup for years > without any problems, but on our new servers with directory container pools > performance is really, really bad. > The servers and storage are designed according to the Blueprints and they > are working fine as long as you do not add large database (Oracle and SAP) > client to them. As soon as you do, the overall server performance becomes > very bad: client and admin session initiation takes 20 to 40 seconds, SQL > queries run for minutes where they should take a few seconds and q query > stgpool sometimes takes more than a minute to respond! > I have two cases open for this. In one case we focused a lot on the OS and > disk performance, but during that process I noticed that the performance is > most likely caused by the way TSM processes large (few hundred MB) files. I > performed a large amount of tests and came to the conclusion that it takes > TSM a huge amount of time to delete large deduplicated files, both in > container pools as deduplicated file pools. As test I use an TDP for Oracle > client which uses a backup piece size of 900 MB. The client contains about > 5000 files. Deleting the files from a container pool takes more than an > hour. When you run a delete object for the files individually I see that > most files take more than a second(!) to delete. If I put that same data in > a non-deduplicated file pool, a delete filespace takes about 15 seconds... > The main
Re: deletion performance of large deduplicated files
Hello Eric, welcome to my nightmares. Take a seat, wanna have a drink? I had the pleasure of performance and data corruption PMRs during the last two years with TDP Oracle. Yes, at first the customer got blamed for not adhering completely to to blueprints, but after some weeks it boild down to ... silence. Data corruption was because of what ended in IT28096 - now fixed. Performance is interesting, but resembles to what you have written. We work with MAXPIECESIZE settings on RMAN to keep the backup pieces small and got some interesting values, pending further observation, but we might be on a cheerful way. I'm talking about database sizes of 50TB here, warehouse style. In between we moved the big DBs to a dedicated server to prove that the performance drop is because of the big DBs, and the remaining "small" DBs - size of 500MB up to 5TB - didn't put any measurable stress on the DB in terms of expiration and protect stgpool. Even the big DBs on their dedicated server performed better in terms of expiration and protect stgpool, which might have been a coincidence of these DBs holding nearly the same data and having the same retention period. What I can't observe is a slowness of the DB. Queries are answered in the normal time - depending on the query. a count(*) from backupobjects naturally takes some time, considerably longer when you use dedup, but the daily queries are answered in the "normal" timeframe. What helped immediately was some tuning: - More LUNS and filesystems for the TSM-DB - smaller disks, but more of them, for each filesystem. changing the disks from 100GB to 2 x 50GB for each DB-filesystem got me a performance boost of 200% in expiration and backup db. Unbelievable, but true. Yes, I'm using SSD. And SVC. And multiple storage systems. Performance isn't the problem, we are measuring 2ms respone time for write AND read. - stripeset for each fileset -- Michael Prix On Fri, 2019-07-19 at 07:29 +, Loon, Eric van (ITOP NS) - KLM wrote: > Hi TSM/SP-ers, > > We are struggling with the performance of our TSM servers for months now. We > are running several servers with hardware (Data Domain) dedup for years > without any problems, but on our new servers with directory container pools > performance is really, really bad. > The servers and storage are designed according to the Blueprints and they > are working fine as long as you do not add large database (Oracle and SAP) > client to them. As soon as you do, the overall server performance becomes > very bad: client and admin session initiation takes 20 to 40 seconds, SQL > queries run for minutes where they should take a few seconds and q query > stgpool sometimes takes more than a minute to respond! > I have two cases open for this. In one case we focused a lot on the OS and > disk performance, but during that process I noticed that the performance is > most likely caused by the way TSM processes large (few hundred MB) files. I > performed a large amount of tests and came to the conclusion that it takes > TSM a huge amount of time to delete large deduplicated files, both in > container pools as deduplicated file pools. As test I use an TDP for Oracle > client which uses a backup piece size of 900 MB. The client contains about > 5000 files. Deleting the files from a container pool takes more than an > hour. When you run a delete object for the files individually I see that > most files take more than a second(!) to delete. If I put that same data in > a non-deduplicated file pool, a delete filespace takes about 15 seconds... > The main issue is that the TDP clients are doing the exact same thing: as > soon as a backup file is no longer needed, it's removed from the RMAN > catalog and deleted from TSM. Since we have several huge database clients > (multiple TB's each) these Oracle delete jobs tend to run for hours. These > delete jobs also seem to slow down each other, because when there are > several of those jobs running at the same time, they become even more slow. > At this point I have one server where these jobs are running 24 hours per > day! This server is at the moment the worst performing TSM server I have > ever seen. On the other container pool servers I was able to move the Oracle > and SAP server away to the old servers (the ones with the Data Domain), but > on this one I can't because of Data Domain capacity reasons. > For this file deletion performance I also have a case open, but there is > absolutely no progress. I proved IBM how bad the performance is and I even > offered them a copy of our database so they can see for themselves, but only > silence from development... > One thing I do not understand: I find it very hard to believe that we are > the only one suffering from this issue. There must be dozens of TSM users > out there that backup large databases to TSM container pools? > > Kind regards, > Eric van Loon > Air France/KLM Storage & Backup > > For
deletion performance of large deduplicated files
Hi TSM/SP-ers, We are struggling with the performance of our TSM servers for months now. We are running several servers with hardware (Data Domain) dedup for years without any problems, but on our new servers with directory container pools performance is really, really bad. The servers and storage are designed according to the Blueprints and they are working fine as long as you do not add large database (Oracle and SAP) client to them. As soon as you do, the overall server performance becomes very bad: client and admin session initiation takes 20 to 40 seconds, SQL queries run for minutes where they should take a few seconds and q query stgpool sometimes takes more than a minute to respond! I have two cases open for this. In one case we focused a lot on the OS and disk performance, but during that process I noticed that the performance is most likely caused by the way TSM processes large (few hundred MB) files. I performed a large amount of tests and came to the conclusion that it takes TSM a huge amount of time to delete large deduplicated files, both in container pools as deduplicated file pools. As test I use an TDP for Oracle client which uses a backup piece size of 900 MB. The client contains about 5000 files. Deleting the files from a container pool takes more than an hour. When you run a delete object for the files individually I see that most files take more than a second(!) to delete. If I put that same data in a non-deduplicated file pool, a delete filespace takes about 15 seconds... The main issue is that the TDP clients are doing the exact same thing: as soon as a backup file is no longer needed, it's removed from the RMAN catalog and deleted from TSM. Since we have several huge database clients (multiple TB's each) these Oracle delete jobs tend to run for hours. These delete jobs also seem to slow down each other, because when there are several of those jobs running at the same time, they become even more slow. At this point I have one server where these jobs are running 24 hours per day! This server is at the moment the worst performing TSM server I have ever seen. On the other container pool servers I was able to move the Oracle and SAP server away to the old servers (the ones with the Data Domain), but on this one I can't because of Data Domain capacity reasons. For this file deletion performance I also have a case open, but there is absolutely no progress. I proved IBM how bad the performance is and I even offered them a copy of our database so they can see for themselves, but only silence from development... One thing I do not understand: I find it very hard to believe that we are the only one suffering from this issue. There must be dozens of TSM users out there that backup large databases to TSM container pools? Kind regards, Eric van Loon Air France/KLM Storage & Backup For information, services and offers, please visit our web site: http://www.klm.com. This e-mail and any attachment may contain confidential and privileged material intended for the addressee only. If you are not the addressee, you are notified that no part of the e-mail or any attachment may be disclosed, copied or distributed, and that any other action related to this e-mail or attachment is strictly prohibited, and may be unlawful. If you have received this e-mail by error, please notify the sender immediately by return e-mail, and delete this message. Koninklijke Luchtvaart Maatschappij NV (KLM), its subsidiaries and/or its employees shall not be liable for the incorrect or incomplete transmission of this e-mail or any attachments, nor responsible for any delay in receipt. Koninklijke Luchtvaart Maatschappij N.V. (also known as KLM Royal Dutch Airlines) is registered in Amstelveen, The Netherlands, with registered number 33014286
